RE: [PATCH net] xen-netback: bookkeep number of queues in our own module

-----Original Message-----
From: David Miller [mailto:davem@davemloft.net]
Sent: 23 June 2014 10:26
To: Paul Durrant
Cc: Wei Liu; xen-devel@lists.xen.org; netdev@vger.kernel.org;
boris.ostrovsky@oracle.com; Ian Campbell
Subject: Re: [PATCH net] xen-netback: bookkeep number of queues in our
own module

From: Paul Durrant <redacted>
Date: Mon, 23 Jun 2014 09:18:23 +0000

quoted

Ok, for the addition of a new algorithm that may be the case but
what about the existent algorithm stability issue? Who's going to
watch the implementation of __netdev_pick_tx() and make sure there
is no semantic change? I'm still of the opinion that the
implementation should be kept within netback so that we can
guarantee stability, even if it means duplication. Either that or we
need some guarantee that the semantic of __netdev_pick_tx() will
never change.

Please stop talking like this.

If the default changes it's probably to fix a bug or make things
better, you want no bugs to get fix if it breaks "stability"?

To some extent, yes. My hypothesis is that the frontend and the backend agree a steering algorithm so that they can ensure all rx and tx traffic for a flow hits a particular CPU, so that locking can be avoided. If the backend then changes its algorithm then packets may arrive on the wrong queue, hit the wrong CPU and now you have a frontend that may try to access critical data structures without locking from two different CPUs. That's going to be a pretty subtle crash to have to debug.

  Paul